DAMM-JSD: A Novel Framework for Handling Missing Values in Concept Drift Detection for Streaming Data

Résumé

Detection of concept drift in streaming data is crucial task especially in the presence of missing values.In this paper a novel approach DAMM-JSD using Jensen-Shannon divergence,which is capable of robustly identifying distributional changes under different mechanisms of missing values such as MCAR,MAR,MNAR. Experimentation has been done on different benchmark datasets using different evaluation metrics and the proposed method DAMM-JSD has performed significanlty better to the existing base methods such as ADWIM and PH. The statistical significance of the method has been validated using Friedman and Nemenyi test. The method has the capability of identifying drift in an incomplete data streams and has greater time complexity compared to base methods.DAMM-JSD provides an robust and accurate solution for detecting concept drift and to address evolving data streams with inherent missing values.